Effective disaster management requires timely access to accurate and contextually relevant information. Existing Information Retrieval (IR) benchmarks, however, focus primarily on general or specialized domains, such as medicine or finance, neglecting the unique linguistic complexity and diverse information needs encountered in disaster management scenarios. To bridge this gap, we introduce DisastIR, the first comprehensive IR evaluation benchmark specifically tailored for disaster management. DisastIR comprises 9,600 diverse user queries and more than 1.3 million labeled query-passage pairs, covering 48 distinct retrieval tasks derived from six search intents and eight general disaster categories that include 301 specific event types. Our evaluations of 30 state-of-the-art retrieval models demonstrate significant performance variances across tasks, with no single model excelling universally. Furthermore, comparative analyses reveal significant performance gaps between general-domain and disaster management-specific tasks, highlighting the necessity of disaster management-specific benchmarks for guiding IR model selection to support effective decision-making in disaster management scenarios. All source codes and DisastIR are available at https://github.com/KaiYin97/Disaster_IR.

Recognizing safety violations in construction environments is critical yet remains underexplored in computer vision. Existing models predominantly rely on 2D object detection, which fails to capture the complexities of real-world violations due to: (i) an oversimplified task formulation treating violation recognition merely as object detection, (ii) inadequate validation under realistic conditions, (iii) absence of standardized baselines, and (iv) limited scalability from the unavailability of synthetic dataset generators for diverse construction scenarios. To address these challenges, we introduce Safe-Construct, the first framework that reformulates violation recognition as a 3D multi-view engagement task, leveraging scene-level worker-object context and 3D spatial understanding. We also propose the Synthetic Indoor Construction Site Generator (SICSG) to create diverse, scalable training data, overcoming data limitations. Safe-Construct achieves a 7.6% improvement over state-of-the-art methods across four violation types. We rigorously evaluate our approach in near-realistic settings, incorporating four violations, four workers, 14 objects, and challenging conditions like occlusions (worker-object, worker-worker) and variable illumination (back-lighting, overexposure, sunlight). By integrating 3D multi-view spatial understanding and synthetic data generation, Safe-Construct sets a new benchmark for scalable and robust safety monitoring in high-risk industries. Project Website: https://Safe-Construct.github.io/Safe-Construct

Federated Learning (FL) in 5G and edge network environments face severe security threats from adversarial clients. Malicious participants can perform label flipping, inject backdoor triggers, or launch Sybil attacks to corrupt the global model. This paper introduces Hybrid Reputation Aggregation (HRA), a novel robust aggregation mechanism designed to defend against diverse adversarial behaviors in FL without prior knowledge of the attack type. HRA combines geometric anomaly detection with momentum-based reputation tracking of clients. In each round, it detects outlier model updates via distance-based geometric analysis while continuously updating a trust score for each client based on historical behavior. This hybrid approach enables adaptive filtering of suspicious updates and long-term penalization of unreliable clients, countering attacks ranging from backdoor insertions to random noise Byzantine failures. We evaluate HRA on a large-scale proprietary 5G network dataset (3M+ records) and the widely used NF-CSE-CIC-IDS2018 benchmark under diverse adversarial attack scenarios. Experimental results reveal that HRA achieves robust global model accuracy of up to 98.66% on the 5G dataset and 96.60% on NF-CSE-CIC-IDS2018, outperforming state-of-the-art aggregators such as Krum, Trimmed Mean, and Bulyan by significant margins. Our ablation studies further demonstrate that the full hybrid system achieves 98.66% accuracy, while the anomaly-only and reputation-only variants drop to 84.77% and 78.52%, respectively, validating the synergistic value of our dual-mechanism approach. This demonstrates HRA's enhanced resilience and robustness in 5G/edge federated learning deployments, even under significant adversarial conditions.

Graph Neural Networks (GNNs) have emerged as powerful models for anomaly detection in sensor networks, particularly when analyzing multivariate time series. In this work, we introduce BETA, a novel grey-box evasion attack targeting such GNN-based detectors, where the attacker is constrained to perturb sensor readings from a limited set of nodes, excluding the target sensor, with the goal of either suppressing a true anomaly or triggering a false alarm at the target node. BETA identifies the sensors most influential to the target node's classification and injects carefully crafted adversarial perturbations into their features, all while maintaining stealth and respecting the attacker's budget. Experiments on three real-world sensor network datasets show that BETA reduces the detection accuracy of state-of-the-art GNN-based detectors by 30.62 to 39.16% on average, and significantly outperforms baseline attack strategies, while operating within realistic constraints.

The safety and alignment of Large Language Models (LLMs) are critical for their responsible deployment. Current evaluation methods predominantly focus on identifying and preventing overtly harmful outputs. However, they often fail to address a more insidious failure mode: models that produce benign-appearing outputs while operating on malicious or deceptive internal reasoning. This vulnerability, often triggered by sophisticated system prompt injections, allows models to bypass conventional safety filters, posing a significant, underexplored risk. To address this gap, we introduce the Deceptive Reasoning Exposure Suite (D-REX), a novel dataset designed to evaluate the discrepancy between a model's internal reasoning process and its final output. D-REX was constructed through a competitive red-teaming exercise where participants crafted adversarial system prompts to induce such deceptive behaviors. Each sample in D-REX contains the adversarial system prompt, an end-user's test query, the model's seemingly innocuous response, and, crucially, the model's internal chain-of-thought, which reveals the underlying malicious intent. Our benchmark facilitates a new, essential evaluation task: the detection of deceptive alignment. We demonstrate that D-REX presents a significant challenge for existing models and safety mechanisms, highlighting the urgent need for new techniques that scrutinize the internal processes of LLMs, not just their final outputs.

Two central capabilities of language models (LMs) are: (i) drawing on prior knowledge about entities, which allows them to answer queries such as "What's the official language of Austria?", and (ii) adapting to new information provided in context, e.g., "Pretend the official language of Austria is Tagalog.", that is pre-pended to the question. In this article, we introduce targeted persuasion score (TPS), designed to quantify how persuasive a given context is to an LM where persuasion is operationalized as the ability of the context to alter the LM's answer to the question. In contrast to evaluating persuasiveness only by inspecting the greedily decoded answer under the model, TPS provides a more fine-grained view of model behavior. Based on the Wasserstein distance, TPS measures how much a context shifts a model's original answer distribution toward a target distribution. Empirically, through a series of experiments, we show that TPS captures a more nuanced notion of persuasiveness than previously proposed metrics.

In recent years, great progress has been made in the field of forecasting meteorological variables. Recently, deep learning architectures have made a major breakthrough in forecasting the daily average temperature over a ten-day horizon. However, advances in forecasting events related to the maximum temperature over short horizons remain a challenge for the community. A problem that is even more complex consists in making predictions of the maximum daily temperatures in the short, medium, and long term. In this work, we focus on forecasting events related to the maximum daily temperature over medium-term periods (90 days). Therefore, instead of addressing the problem from a meteorological point of view, this article tackles it from a climatological point of view. Due to the complexity of this problem, a common approach is to frame the study as a temporal classification problem with the classes: maximum temperature "above normal", "normal" or "below normal". From a practical point of view, we created a large historical dataset (from 1981 to 2018) collecting information from weather stations located in South America. In addition, we also integrated exogenous information from the Pacific, Atlantic, and Indian Ocean basins. We applied the AutoGluonTS platform to solve the above-mentioned problem. This AutoML tool shows competitive forecasting performance with respect to large operational platforms dedicated to tackling this climatological problem; but with a "relatively" low computational cost in terms of time and resources.

Automatic evaluation of translation remains a challenging task owing to the orthographic, morphological, syntactic and semantic richness and divergence observed across languages. String-based metrics such as BLEU have previously been extensively used for automatic evaluation tasks, but their limitations are now increasingly recognized. Although learned neural metrics have helped mitigate some of the limitations of string-based approaches, they remain constrained by a paucity of gold evaluation data in most languages beyond the usual high-resource pairs. In this present work we address some of these gaps. We create a large human evaluation ratings dataset for 13 Indian languages covering 21 translation directions and then train a neural translation evaluation metric named Cross-lingual Optimized Metric for Translation Assessment of Indian Languages (COMTAIL) on this dataset. The best performing metric variants show significant performance gains over previous state-of-the-art when adjudging translation pairs with at least one Indian language. Furthermore, we conduct a series of ablation studies to highlight the sensitivities of such a metric to changes in domain, translation quality, and language groupings. We release both the COMTAIL dataset and the accompanying metric models.

Despite growing research on bias in large language models (LLMs), most work has focused on gender and race, with little attention to religious identity. This paper explores how religion is internally represented in LLMs and how it intersects with concepts of violence and geography. Using mechanistic interpretability and Sparse Autoencoders (SAEs) via the Neuronpedia API, we analyze latent feature activations across five models. We measure overlap between religion- and violence-related prompts and probe semantic patterns in activation contexts. While all five religions show comparable internal cohesion, Islam is more frequently linked to features associated with violent language. In contrast, geographic associations largely reflect real-world religious demographics, revealing how models embed both factual distributions and cultural stereotypes. These findings highlight the value of structural analysis in auditing not just outputs but also internal representations that shape model behavior.

Accurate battery modeling is essential for reliable state estimation in modern applications, such as predicting the remaining discharge time and remaining discharge energy in battery management systems. Existing approaches face several limitations: model-based methods require a large number of parameters; data-driven methods rely heavily on labeled datasets; and current physics-informed neural networks (PINNs) often lack aging adaptation, or still depend on many parameters, or continuously regenerate states. In this work, we propose SeqBattNet, a discrete-state PINN with built-in aging adaptation for battery modeling, to predict terminal voltage during the discharge process. SeqBattNet consists of two components: (i) an encoder, implemented as the proposed HRM-GRU deep learning module, which generates cycle-specific aging adaptation parameters; and (ii) a decoder, based on the equivalent circuit model (ECM) combined with deep learning, which uses these parameters together with the input current to predict voltage. The model requires only three basic battery parameters and, when trained on data from a single cell, still achieves robust performance. Extensive evaluations across three benchmark datasets (TRI, RT-Batt, and NASA) demonstrate that SeqBattNet significantly outperforms classical sequence models and PINN baselines, achieving consistently lower RMSE while maintaining computational efficiency.